Skip to content

IQ2_K_R4 #146

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Dec 17, 2024
Merged

IQ2_K_R4 #146

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
d5d94bb
iq2_k_r4: Zen4
Kawrakow Dec 17, 2024
bb4cc44
iq2_k_r4: NEON
Kawrakow Dec 17, 2024
a75a790
iq2_k_r4: better matrix x vector multiplication on NEON
Kawrakow Dec 17, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions examples/quantize/quantize.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,7 @@ static const std::vector<struct quant_option> QUANT_OPTIONS = {
{ "IQ4_KS", LLAMA_FTYPE_MOSTLY_IQ4_KS, " 4.25 bpw non-linear quantization", },
{ "IQ4_KSS", LLAMA_FTYPE_MOSTLY_IQ4_KSS, " 4.0 bpw non-linear quantization", },
{ "IQ2_K", LLAMA_FTYPE_MOSTLY_IQ2_K, " 2.375 bpw non-linear quantization",},
{ "IQ2_K_R4", LLAMA_FTYPE_MOSTLY_IQ2_K_R4, "IQ2_K repacked",},
{ "IQ2_KS", LLAMA_FTYPE_MOSTLY_IQ2_KS, " 2.1875 bpw non-linear quantization",},
{ "IQ3_K", LLAMA_FTYPE_MOSTLY_IQ3_K, " 3.44 bpw non-linear quantization", },
{ "IQ3_K_R4", LLAMA_FTYPE_MOSTLY_IQ3_K_R4, "IQ3_K repacked", },
Expand Down
2 changes: 2 additions & 0 deletions ggml/include/ggml.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -423,6 +423,7 @@ extern "C" {
GGML_TYPE_BF16_R16 = 230,
GGML_TYPE_Q6_0_R4 = 233,
GGML_TYPE_IQ2_BN_R4 = 335,
GGML_TYPE_IQ2_K_R4 = 337,
GGML_TYPE_IQ3_K_R4 = 338,
GGML_TYPE_IQ4_K_R4 = 339,
GGML_TYPE_Q8_K_R8 = 399,
Expand Down Expand Up @@ -498,6 +499,7 @@ extern "C" {
GGML_FTYPE_MOSTLY_BF16_R16 = 224, // except 1d tensors
GGML_FTYPE_MOSTLY_Q6_0_R4 = 227, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ2_BN_R4 = 329, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ2_K_R4 = 330, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ3_K_R4 = 331, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ4_K_R4 = 332, // except 1d tensors
GGML_FTYPE_MOSTLY_Q8_K_R8 = 399, // except 1d tensors
Expand Down
8 changes: 8 additions & 0 deletions ggml/src/ggml-common.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -521,6 +521,14 @@ typedef struct {
} block_iq2_k;
static_assert(sizeof(block_iq2_k) == sizeof(ggml_half) + sizeof(uint16_t) + QK_K/32 + QK_K/4, "wrong iq2_k block size/padding");

typedef struct {
ggml_half d[4];
uint8_t extra[8];
uint8_t scales[QK_K/8];
uint8_t qs[QK_K];
} block_iq2_k_r4;
static_assert(sizeof(block_iq2_k_r4) == 4*sizeof(block_iq2_k), "wrong iq2_k_r4 block size/padding");

typedef struct {
uint16_t extra;
uint8_t scales[QK_K/64];
Expand Down
1 change: 1 addition & 0 deletions ggml/src/ggml-quants.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15207,6 +15207,7 @@ bool ggml_validate_row_data(enum ggml_type type, const void * data, size_t nbyte
case GGML_TYPE_Q4_K_R4: break;
case GGML_TYPE_Q5_K_R4: break;
case GGML_TYPE_Q6_K_R4: break;
case GGML_TYPE_IQ2_K_R4: break;
case GGML_TYPE_IQ3_K_R4: break;
case GGML_TYPE_IQ4_K_R4: break;
case GGML_TYPE_Q8_K_R8: break;
Expand Down
22 changes: 22 additions & 0 deletions ggml/src/ggml.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1308,6 +1308,19 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
.nrows = 1,
.row_meta_size = 0,
},
[GGML_TYPE_IQ2_K_R4] = {
.type_name = "iq2_k_r4",
.blck_size = QK_K,
.type_size = sizeof(block_iq2_k),
.is_quantized = true,
.to_float = (ggml_to_float_t) dequantize_row_iq2_k_r4,
.from_float = quantize_row_iq2_k_r4,
.from_float_ref = (ggml_from_float_t)quantize_row_iq2_k_r4_ref,
.vec_dot = vec_dot_iq2_k_r4_q8_k,
.vec_dot_type = GGML_TYPE_Q8_K,
.nrows = 1,
.row_meta_size = 0,
},
[GGML_TYPE_IQ2_KS] = {
.type_name = "iq2_ks",
.blck_size = QK_K,
Expand Down Expand Up @@ -4173,6 +4186,7 @@ enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype) {
case GGML_FTYPE_MOSTLY_IQ4_KS: wtype = GGML_TYPE_IQ4_KS; break;
case GGML_FTYPE_MOSTLY_IQ4_KSS: wtype = GGML_TYPE_IQ4_KSS; break;
case GGML_FTYPE_MOSTLY_IQ2_K: wtype = GGML_TYPE_IQ2_K; break;
case GGML_FTYPE_MOSTLY_IQ2_K_R4: wtype = GGML_TYPE_IQ2_K_R4; break;
case GGML_FTYPE_MOSTLY_IQ2_KS: wtype = GGML_TYPE_IQ2_KS; break;
case GGML_FTYPE_MOSTLY_IQ3_K: wtype = GGML_TYPE_IQ3_K; break;
case GGML_FTYPE_MOSTLY_IQ4_K: wtype = GGML_TYPE_IQ4_K; break;
Expand Down Expand Up @@ -10711,6 +10725,7 @@ static void ggml_compute_forward_add(
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ2_K:
case GGML_TYPE_IQ2_K_R4:
case GGML_TYPE_IQ2_KS:
case GGML_TYPE_IQ3_K:
case GGML_TYPE_IQ4_K:
Expand Down Expand Up @@ -11168,6 +11183,7 @@ static void ggml_compute_forward_add1(
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ2_K:
case GGML_TYPE_IQ2_K_R4:
case GGML_TYPE_IQ2_KS:
case GGML_TYPE_IQ3_K:
case GGML_TYPE_IQ4_K:
Expand Down Expand Up @@ -11322,6 +11338,7 @@ static void ggml_compute_forward_acc(
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ2_K:
case GGML_TYPE_IQ2_K_R4:
case GGML_TYPE_IQ2_KS:
case GGML_TYPE_IQ3_K:
case GGML_TYPE_IQ4_K:
Expand Down Expand Up @@ -14522,6 +14539,7 @@ static void ggml_compute_forward_out_prod(
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ2_K:
case GGML_TYPE_IQ2_K_R4:
case GGML_TYPE_IQ2_KS:
case GGML_TYPE_IQ3_K:
case GGML_TYPE_IQ4_K:
Expand Down Expand Up @@ -14916,6 +14934,7 @@ static void ggml_compute_forward_set(
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ2_K:
case GGML_TYPE_IQ2_K_R4:
case GGML_TYPE_IQ2_KS:
case GGML_TYPE_IQ3_K:
case GGML_TYPE_IQ4_K:
Expand Down Expand Up @@ -15204,6 +15223,7 @@ static void ggml_compute_forward_get_rows(
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ2_K:
case GGML_TYPE_IQ2_K_R4:
case GGML_TYPE_IQ2_KS:
case GGML_TYPE_IQ3_K:
case GGML_TYPE_IQ4_K:
Expand Down Expand Up @@ -15821,6 +15841,7 @@ static void ggml_compute_forward_clamp(
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ2_K:
case GGML_TYPE_IQ2_K_R4:
case GGML_TYPE_IQ2_KS:
case GGML_TYPE_IQ3_K:
case GGML_TYPE_IQ4_K:
Expand Down Expand Up @@ -22666,6 +22687,7 @@ size_t ggml_quantize_chunk(
case GGML_TYPE_IQ4_KS: result = quantize_iq4_ks (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ4_KSS: result = quantize_iq4_kss(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ2_K: result = quantize_iq2_k (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ2_K_R4:result = quantize_iq2_k_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ2_KS: result = quantize_iq2_ks (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ3_K: result = quantize_iq3_k (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ4_K: result = quantize_iq4_k (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
Expand Down
Loading